Model playground

Browser-native transformer inference. WebGPU-accelerated, no server required. Models load directly from Hugging Face.

Image Classification ViT-B/16

Classify an image into 1000 ImageNet categories.

85 MB

Drop an image here or click to browse

Zero-Shot Classification CLIP ViT-B/32

Match an image against your own labels — no training required.

150 MB

Drop an image here or click to browse

Fast Chat LFM2.5 350M

Official LFM2.5 350M — same size, newer architecture.

~276 MB

Text Generation LFM2.5 1.2B

Chat with a Liquid Foundation Model. Runs fully in-browser.

1.2 GB

Thinking LFM2.5 1.2B Thinking

Dedicated reasoning model — shows its thinking steps before answering.

1.2 GB

Tool Calling LFM2.5 1.2B WebMCP

Model calls browser-registered tools via navigator.modelContext.

shared